• Taobao, as the largest online retail platform in the world, provides billions of online display advertising impressions for millions of advertisers every day. For commercial purposes, the advertisers bid for specific spots and target crowds to compete for business traffic. The platform chooses the most suitable ads to display in tens of milliseconds. Common pricing methods include cost per mille (CPM) and cost per click (CPC). Traditional advertising systems target certain traits of users and ad placements with fixed bids, essentially regarded as coarse-grained matching of bid and traffic quality. However, the fixed bids set by the advertisers competing for different quality requests cannot fully optimize the advertisers' key requirements. Moreover, the platform has to be responsible for the business revenue and user experience. Thus, we proposed a bid optimizing strategy called optimized cost per click (OCPC) which automatically adjusts the bid to achieve finer matching of bid and traffic quality of page view (PV) request granularity. Our approach optimizes advertisers' demands, platform business revenue and user experience and as a whole improves traffic allocation efficiency. We have validated our approach in Taobao display advertising system in production. The online A/B test shows our algorithm yields substantially better results than previous fixed bid manner.
  • Click-through rate prediction is an essential task in industrial applications, such as online advertising. Recently deep learning based models have been proposed, which follow a similar Embedding\&MLP paradigm. In these methods large scale sparse input features are first mapped into low dimensional embedding vectors, and then transformed into fixed-length vectors in a group-wise manner, finally concatenated together to fed into a multilayer perceptron (MLP) to learn the nonlinear relations among features. In this way, user features are compressed into a fixed-length representation vector, in regardless of what candidate ads are. The use of fixed-length vector will be a bottleneck, which brings difficulty for Embedding\&MLP methods to capture user's diverse interests effectively from rich historical behaviors. In this paper, we propose a novel model: Deep Interest Network (DIN) which tackles this challenge by designing a local activation unit to adaptively learn the representation of user interests from historical behaviors with respect to a certain ad. This representation vector varies over different ads, improving the expressive ability of model greatly. Besides, we develop two techniques: mini-batch aware regularization and data adaptive activation function which can help training industrial deep networks with hundreds of millions of parameters. Experiments on two public datasets as well as an Alibaba real production dataset with over 2 billion samples demonstrate the effectiveness of proposed approaches, which achieve superior performance compared with state-of-the-art methods. DIN now has been successfully deployed in the online display advertising system in Alibaba, serving the main traffic.
  • Model-based methods for recommender systems have been studied to provide more precise results. In systems with large corpus, the amount of calculation for learnt model to predict all user-item pairs' preferences is tremendous, which makes the model difficult to be directly employed in recommendation candidate generation stage. To overcome the calculation barrier, models like matrix factorization can resort to inner product form (i.e., use the inner product of user and item's latent factors as the preference) and index like hashing to perform efficient approximate k-nearest neighbor search. However, other more expressive interaction forms between user and item features, e.g., interactions through advanced deep neural networks, are still prevented from large corpus recommendation because of the amount of calculation. In this paper, we focus on the problem how arbitrary advanced models can be introduced to generate recommendations from large corpus. We propose a novel tree-based method which can provide logarithmic complexity prediction w.r.t. corpus size with more expressive deep neural networks. The main idea of tree-based model is to predict user interests coarse-to-fine, by traversing tree nodes top-down and making decisions whether to pick up each node to user. Furthermore, we show that the tree structure can also be jointly learnt towards better compatible with user interests' distribution, to facilitate both training and prediction. Experiments in two large-scale real-world datasets indicate that the proposed model significantly outperforms traditional methods. And online A/B test results in Taobao display advertising platform prove the effectiveness of the tree-based deep model in production.
  • Deep networks have been successfully applied to learn transferable features for adapting models from a source domain to a different target domain. In this paper, we present joint adaptation networks (JAN), which learn a transfer network by aligning the joint distributions of multiple domain-specific layers across domains based on a joint maximum mean discrepancy (JMMD) criterion. Adversarial training strategy is adopted to maximize JMMD such that the distributions of the source and target domains are made more distinguishable. Learning can be performed by stochastic gradient descent with the gradients computed by back-propagation in linear-time. Experiments testify that our model yields state of the art results on standard datasets.
  • The recent success of deep neural networks relies on massive amounts of labeled data. For a target task where labeled data is unavailable, domain adaptation can transfer a learner from a different source domain. In this paper, we propose a new approach to domain adaptation in deep networks that can jointly learn adaptive classifiers and transferable features from labeled data in the source domain and unlabeled data in the target domain. We relax a shared-classifier assumption made by previous methods and assume that the source classifier and target classifier differ by a residual function. We enable classifier adaptation by plugging several layers into deep network to explicitly learn the residual function with reference to the target classifier. We fuse features of multiple layers with tensor product and embed them into reproducing kernel Hilbert spaces to match distributions for feature adaptation. The adaptation can be achieved in most feed-forward models by extending them with new residual layers and loss functions, which can be trained efficiently via back-propagation. Empirical evidence shows that the new approach outperforms state of the art methods on standard domain adaptation benchmarks.
  • Aug. 1, 2015 cs.LO
    Name-passing calculi are foundational models for mobile computing. Research into these models has produced a wealth of results ranging from relative expressiveness to programming pragmatics. The diversity of these results call for clarification and reorganization. This paper applies a model independent approach to the study of the name-passing calculi, leading to a uniform treatment and simplification. The technical tools and the results presented in the paper form the foundation for a theory of name-passing calculus.
  • We report studies of pinning mode resonances of magnetic field induced bilayer Wigner crystals of bilayer hole samples with negligible interlayer tunneling and different interlayer separations d, in states with varying layer densities, including unequal layer densities. With unequal layer densities, samples with large d relative to the in-plane carrier-carrier spacing a, two pinning resonances are present, one for each layer. For small d/a samples, a single resonance is observed even with significant density imbalance. These samples, at balance, were shown to exhibit an enhanced pinning mode frequency [Zhihai Wang et al., Phys. Rev. Lett. 136804 (2007)], which was ascribed to a one-component, pseudospin ferromagnetic Wigner solid. The evolution of the resonance frequency and line width indicates the quantum interlayer coherence survives at moderate density imbalance, but disappears when imbalance is sufficiently large.
  • We report the observation of a resonance in the microwave spectra of the real diagonal conductivities of a two-dimensional electron system within a range of ~ +- .0.015 $ from filling factor $\nu=1/3$. The resonance is remarkably similar to resonances previously observed near integer $\nu$, and is interpreted as the collective pinning mode of a disorder-pinned Wigner solid phase of $e/3$-charged carriers .
  • Microwave pinning-mode resonances found around integer quantum Hall effects, are a signature of crystallized quasiparticles or holes. Application of in-plane magnetic field to these crystals, increasing the Zeeman energy, has negligible effect on the resonances just below Landau level filling $\nu=2$, but increases the pinning frequencies near $\nu=1$, particularly for smaller quasiparticle/hole densities. The charge dynamics near $\nu=1$, characteristic of a crystal order, are affected by spin, in a manner consistent with a Skyrme crystal.
  • We study the anisotropic pinning-mode resonances in the rf conductivity spectra of the stripe phase of 2D electron systems (2DES) around Landau level filling 9/2, in the presence of an in-plane magnetic field, B_ip. The polarization along which the resonance is observed switches as B_ip is applied, consistent with the reorientation of the stripes. The resonance frequency, a measure of the pinning interaction between the 2DES and disorder, increases with B_ip. The magnitude of this increase indicates that disorder interaction is playing an important role in determining the stripe orientation.
  • We study the radio-frequency diagonal conductivities of the anisotropic stripe phases of higher Landau levels near half integer fillings. In the hard direction, in which larger dc resistivity occurs, the spectrum exhibits a striking resonance, while in the orthogonal, easy direction, no resonance is discernable. The resonance is interpreted as a pinning mode of the stripe phase.
  • We propose a dielectrophoresis model for phase-separated manganites. Without increase of the fraction of metallic phase, an insulator-metal transition occurs when a uniform electric field applied across the system exceeds a threshold value. Driven by the dielectrophoretic force, the metallic clusters reconfigure themselves into stripes along the direction of electric field, leading to the filamentous percolation. This process, which is time-dependent, irreversible and anisotropic, is a probable origin of the colossal electroresistance in manganites.
  • Current theoretical approaches to manganites mainly stem from magnetic framework, in which the electronic transport is thought to be spin-dependent and the double exchange mechanism plays a core role. However, quite a number of experimental observations can yet not be reasonably explained. For example, multiplicate insulator-metal transitions and resistivity reduction induced by perturbations other than magnetic field, such as electric current, are not well understood. A comprehensive analysis on earlier extensive studies is performed and two types of origins for resistivity change are highlighted. Besides the insulated-to-metallic transition induced by external field such as magnetic field, the insulated-to-insulated transition induced extrinsically is even a more important source for the colossal resistivity change. We propose an extended framework for the electronic transport of manganites, in which the contribution of charge degree of freedom is given a special priority.
  • It is commonly known that there exist short paths between vertices in a network showing the small-world effect. Yet vertices, for example, the individuals living in society, usually are not able to find the shortest paths, due to the very serious limit of information. To theoretically study this issue, here the navigation process of launching messages toward designated targets is investigated on a variant of the one-dimensional small-world network (SWN). In the network structure considered, the probability of a shortcut falling between a pair of nodes is proportional to $r^{-\alpha}$, where $r$ is the lattice distance between the nodes. When $\alpha =0$, it reduces to the SWN model with random shortcuts. The system shows the dynamic small-world (SW) effect, which is different from the well-studied static SW effect. We study the effective network diameter, the path length as a function of the lattice distance, and the dynamics. They are controlled by multiple parameters, and we use data collapse to show that the parameters are correlated. The central finding is that, in the one-dimensional network studied, the dynamic SW effect exists for $0\leq \alpha \leq 2$. For each given value of $\alpha $ in this region, the point that the dynamic SW effect arises is $ML^{\prime}\sim 1$, where $M$ is the number of useful shortcuts and $L^{\prime}$ is the average reduced (effective) length of them.
  • The hysteresis of the Ising model in a spatially homogeneous AC field is studied using both mean-field calculations and two-dimensional Monte Carlo simulations. The frequency dispersion and the temperature dependence of the hysteresis loop area are studied in relation to the dynamic symmetry loss. The dynamic mechanisms may be different when the hysteresis loops are symmetric or asymmetric, and they can lead to a double-peak frequency dispersion and qualitatively different temperature dependence.
  • In this article, we investigate the competing Glauber-type and Kawasaki-type dynamics with small-world network (SWN) effect, in the framework of the Gaussian model. The Glauber-type single-spin transition mechanism with probability p simulates the contact of the system with a heat bath and the Kawasaki-type dynamics with probability 1-p simulates an external energy flux. Two different types of SWN effect are studied, one with the total number of links increased and the other with it conserved. The competition of the dynamics leads to an interesting self-organization process that can be characterized by a phase diagram with two identifiable temperatures. By studying the modification of the phase diagrams, the SWN effect on the two dynamics is analyzed. For the Glauber-type dynamics, more important is the altered average coordination number while the Kawasaki-type dynamics is enhanced by the long range spin interaction and redistribution.
  • In network evolution, the effect of aging is universal: in scientific collaboration network, scientists have a finite time span of being active; in movie actors network, once popular stars are retiring from stage; devices on the Internet may become outmoded with techniques developing so rapidly. Here we find in citation networks that this effect can be represented by an exponential decay factor, $e^{-\beta \tau}$, where $\tau $ is the node age, while other evolving networks (the Internet for instance) may have different types of aging, for example, a power-law decay factor, which is also studied and compared. It has been found that as soon as such a factor is introduced to the Barabasi-Albert Scale-Free model, the network will be significantly transformed. The network will be clustered even with infinitely large size, and the clustering coefficient varies greatly with the intensity of the aging effect, i.e. it increases linearly with $\beta $ for small values of $\beta $ and decays exponentially for large values of $\beta $. At the same time, the aging effect may also result in a hierarchical structure and a disassortative degree-degree correlation. Generally the aging effect will increase the average distance between nodes, but the result depends on the type of the decay factor. The network appears like a one-dimensional chain when exponential decay is chosen, but with power-law decay, a transformation process is observed, i.e., from a small-world network to a hypercubic lattice, and to a one-dimensional chain finally. The disparities observed for different choices of the decay factor, in clustering, average node distance and probably other aspects not yet identified, are believed to bear significant meaning on empirical data acquisition.
  • In spin-lattice models with order parameter conserved, we generalize the idea of spin diffusion incorporating a variety factors as possible driving forces, including the external field and the temperature. The Kawasaki dynamics in the Gaussian model and the one-dimensional Ising model are studied. The Gaussian model is rigorously treated and the critical exponent $\gamma =1$ is obtained. The competition of the internal and the external inhomogeneities may lead to interesting and rich dynamic behavior. The diffusion induced by the inhomogeneity of the magnetization itself is believed to vanish near the critical point, meanwhile the nonvanishing diffusion induced by the inhomogeneity of the environment may be coupled to the spin configuration and weakened by thermal noise. Several interesting examples are visualized, and the concept of local hysteresis is proposed in this spin-conserved dynamics. A dynamic phase transition is observed in the one-dimensional Ising model subject to an electromagnetic wave.
  • We analytically investigate the kinetic Gaussian model and the one-dimensional kinetic Ising model on two typical small-world networks (SWN), the adding-type and the rewiring-type. The general approaches and some basic equations are systematically formulated. The rigorous investigation of the Glauber-type kinetic Gaussian model shows the mean-field-like global influence on the dynamic evolution of the individual spins. Accordingly a simplified method is presented and tested, and believed to be a good choice for the mean-field transition widely (in fact, without exception so far) observed on SWN. It yields the evolving equation of the Kawasaki-type Gaussian model. In the one-dimensional Ising model, the p-dependence of the critical point is analytically obtained and the inexistence of such a threshold p_c, for a finite temperature transition, is confirmed. The static critical exponents, gamma and beta are in accordance with the results of the recent Monte Carlo simulations, and also with the mean-field critical behavior of the system. We also prove that the SWN effect does not change the dynamic critical exponent, z=2, for this model. The observed influence of the long-range randomness on the critical point indicates two obviously different hidden mechanisms.
  • In this article, we retain the basic idea and at the same time generalize Kawasaki's dynamics, spin-pair exchange mechanism, to spin-pair redistribution mechanism, and present a normalized redistribution probability. This serves to unite various order-parameter-conserved processes in microscopic, place them under the control of a universal mechanism and provide the basis for further treatment. As an example of the applications, we treated the kinetic Gaussian model and obtained exact diffusion equation. We observed critical slowing down near the critical point and found that, the critical dynamic exponent z=1/nu=2 is independent of space dimensionality and the assumed mechanism, whether Glauber-type or Kawasaki-type.
  • In this article, we have given a systematic formulation of the new generalized competing mechanism: the Glauber-type single-spin transition mechanism, with probability p, simulates the contact of the system with the heat bath, and the Kawasaki-type spin-pair redistribution mechanism, with probability 1-p, simulates an external energy flux. These two mechanisms are natural generalizations of Glauber's single-spin flipping mechanism and Kawasaki's spin-pair exchange mechanism respectively. On the one hand, the new mechanism is in principle applicable to arbitrary systems, while on the other hand, our formulation is able to contain a mechanism that just directly combines single-spin flipping and spin-pair exchange in their original form. Compared with the conventional mechanism, the new mechanism does not assume the simplified version and leads to greater influence of temperature. The fact, order for lower temperature and disorder for higher temperature, will be universally true. In order to exemplify this difference, we applied the mechanism to 1D Ising model and obtained analytical results. We also applied this mechanism to kinetic Gaussian model and found that, above the critical point there will be only paramagnetic phase, while below the critical point, the self-organization as a result of the energy flux will lead the system to an interesting heterophase, instead of the initially guessed antiferromagnetic phase. We studied this process in details.